Results: Adobe Creative Cloud
We have a long-standing issue with our After Effects test, where it doesn't respond positively to more than four processing cores. In fact, we've seen instances where adding more compute horsepower actually yields a negative effect.
As you can see, with nine cores (the maximum we can allocate on the Precision T5600 workstation due to its 16 GB of memory) the workload takes two seconds longer to complete than with four cores. Hours have gone into troubleshooting this since Chris first noticed it, and I finally figured out the source of the issue: QuickTime. The source for the three window overlays are QuickTime footage encoded using the QuickTime PNG codec, making them effectively lossless. So, I changed them out for PNG image sequences.
The result is a slight improvement on the baseline machine, but a huge improvement favoring Dell's Precision T5600. It literally renders the test in one-third of the time using the same nine cores.
The side effect, of course, is that the After Effects motion graphics test we've been using to benchmark is finishing too fast on the latest hardware. I decided to increase the workload's demands by upping the project to 1080p. This was achieved by using Illustrator to convert the background graphic into a structured drawing, and then resizing and repositioning the video overlays for a 16:9 aspect ratio.
This was a good idea at first. And it worked great until I moved the test from the baseline machine's hard drive to an SSD, at which point the complete time went from over two minutes to 71 seconds. Then I ran it on the Dell workstation, which chewed through the task in 24 seconds. That's just one extra second more than the 720p-based benchmark, even though the frame being rendered is six times larger.
On the other hand, the HD version is now much more sensitive to storage performance. That's not a bad thing, given how hard it can be to demonstrate the real-world performance benefits of SSDs. Given more system memory, the T5600 could have utilized its additional execution cores, and the test might have finished even faster.
This test involves our standard Hollywood Sequence project featuring DVCProHD 720p footage. Since the footage is only 100 Mb/s, it isn't as sensitive to storage performance. And as a result of optimizations for GPU acceleration, it's much more affected by graphics performance. Remember, this is the latest version of Premiere Pro, so it is OpenCL-enabled, so it also works with AMD's cards.
The encode to H.264, however, is still an entirely CPU-based affair, which is more taxing. Previously, I was rendering transitions and encoding separately so that both tasks could be evaluated on their own. But to make my tests consistent with Premiere benchmarks on other hardware, I used the same settings. Still, the Dell proves itself just under three times faster than our baseline.
Our two Photoshop tests are based on a series of filters applied to a very large image. They were selected for their prodigious use of threading (in the case of the CPU test) and OpenCL acceleration (for that corresponding benchmark).
The CPU-based Photoshop test falls right about where we'd expect it, just under three times faster over the baseline machine. But the OpenCL result is more difficult to explain. This behavior shows up in our WinZip tests, too. I tried a couple of different Quadro driver revisions and saw little or no difference.